DIPS

Even if BABIP were totally random, there would still be a sizable spread between the highest and lowest observed result in a sample as large as "every MLB pitcher ever." And the spread isn't that big. Amongst the 220 pitchers with 1000+ IP during the Wild Card era (since 1994), the highest career BABIP is .326 and the lowest is .251 (with only four guys below .270). A 10% spread in either direction of the mean is pretty much nothing.

The high school thing is irrelevant-- by the time a pitcher reaches the high minors, the talent pool has already been significantly narrowed. At best, it's an argument that BABIP regression is less useful in the low minors, which I would agree with and isn't at issue here.

I've never said BABIP isn't a skill. I've said that (a) a pitcher's BABIP, especially in small samples, is much more likely to be driven by chance than skill, (b) we know most of the attributes that can drive the ability to sustain a better-than-average BABIP, and (c) even those pitchers who can sustainably outperform the league-average BABIP can only do so by a little bit.

In light of the above, I'm not very impressed by any five- or ten-game stretch where a pitcher (especially one in the high minors) puts up a great ERA on the basis of an exceedingly low BABIP and middling walk/strikout rates. Hell, I'm not that impressed by any season which is driven by an unsustainably low BABIP. It also means that using SIERA or even good old FIP to project pitchers will be a lot more accurate than doing so with ERA, even for minor league pitchers. The spread in true-talent BABIP is just small enough that in a small (read: non-multi-season) sample, it is much more likely to be random (unsustainable) than skill (sustainable).

Relief pitchers are an entirely different animal. Opposing hitters only to see them once (which matters a lot) and they don't have to pace themselves. Many elite relievers were formerly middling starting pitchers (including Koji himself, but also think Wade Davis, Dellin Betances, Andrew Miller, even Mariano Rivera), which suggests that comparing relievers and starters is not usually the best way to go (there's also the fact that they pitch so few innings that it's harder to weed out SSS stuff).

Relative to the league-average in their era, Nolan Ryan and Sandy Koufax are two of the highest strikeout pitchers ever. So yeah, that's no coincidence. You're giving short shift to how important striking guys out and not walking them are. Koji literally struck out more than twice as many hitters as Doubront did this year while walking a third as many-- that goes a long way in explaining their divergent results.

When pitchers are injured or just really bad and thus perform at a sub-MLB caliber, yeah, they can put up exaggerated BABIPs (though I also think we sometimes have the causation backwards-- we only think guys are "truly terrible" as opposed to run-of-the-mill bad when they give up inflated results on BIP). After all, one of the main drivers for DIPS is survivor bias. But that's not relevant to the core of the discussion here, which is how best to evaluate guys putting up streaks of unsustainably low BABIPs.

I'm not a trained scout and I don't get to watch as many games as I'd like. As such, I can either evaluate someone based on their ERA or based on their advanced stats, and it's no question which is the more predictive metric, especially over small samples. More importantly, it is very difficult to sustainably generate "weak contact" without getting strikeouts. Either your stuff is good enough that you start getting more swings-and-misses, or it's not and balls will start falling and/or getting hit harder.

...and his strikeout rate jumped from 6.48 K/9 in 66.2 IP as a starter to 11.53 K/9 as a reliever. Perhaps more relevantly, he put up a .302 BABIP as a starter versus .229 as a reliever (albeit in just 278.2 relief innings).The point is you can't really compare starters with relievers.

Not another dips thread. No, minor-league players do not have special skills that major-league players do not have. It's a selection issue. The only issue is the spread of the demonstrated skill set amongst the population, not if the population has any skills at all. Okay, now that this is done, back to discussing a major-league position player rather than minor-league pictures in general.

If Pedro Martinez pitched in Division 3 college ball when he was 25, he would have better BABIP skills than a batting practice pitcher.  That's an extreme example, because apparently it takes one to get people to understand that some pitchers don't give up much hard contact and worse ones give up a lot more.  And if you put Mookie Betts at pitcher, he'd probably have a higher BABIP because he'd get hit harder.  There, on topic.

That's why I mentioned "except in the minors" where the talent differences are greater.

It's pretty easy to see it when it happens.  A pitcher who gives up 10 runs on 8 infield singles and a couple errors is having bad BABIP luck.  A pitcher who is getting absolutely pounded with line drives is not having bad BABIP luck.  If that happens for a length of time, his BABIP will not return to average and it will remain high.

A simple example is John Lackey in 2011.  He was terrible, pitching with a torn UCL.  His BABIP was high at .339, so his FIP was 1.7 runs lower than his ERA.  But it was because he was terrible, not because he was unlucky.

They don't continue pitching (or they get TJ surgery), they don't make the majors, and they drop out of DIPS studies/the population. Really, everything you are writing has been written 10,000,000 times on the internet before.

Maybe we can get Eric on this... Each hitter's BABIP a normal random variate year over year, with the mean of that variate a function of the line drive rate. That particular stochastic process describes what's going on I think.

I do not buy it, sorry.  If an extreme difference in talent can explain an extreme BABIP difference (high school vs. HOF pitcher), it's logical to believe a smaller difference in talent can also result in a smaller difference.  That difference cannot just disappear because someone says that all major league pitchers are equal and the only thing that separates them is K/BB rates.  I mean hell, overall BABIP doesn't even take GB/FB/LD rates into consideration.  Given the fairly significant differences in BA for different contact types, if you don't believe BABIP is a skill, I'm not sure you can believe that a pitcher can be a ground ball pitcher either.

Note the discussion above about useful simplifications. By the point that pitchers reach the majors, the true-talent difference is so small and the degree of random fluctuation inherent in the stat is such that best way to evaluate pitchers includes assuming that he will regress to a league-average BABIP unless he possesses one of a small number of known skills that are linked to the ability to sustain a better-than-league-average BABIP. The reason GB/FB rates don't affect BABIP skill that much are that the BABIPs for ground balls and outfield fly balls are fairly similar. LD rates require an extremely large sample size before they stabilize, so that's hard to evaluate before you have 3-5 seasons of data.

except there is not much evidence at all that suggests pitchers have control over their BABIP, so to suggest this means you have either found some sabermetric breakthrough or just making stuff up

I haven't come up with a list that shows that is survival bias. Non-MLB pitchers probably leave the league all the time with ERAs much higher than their FIPs.

Personally, I can't wait until there are new stats using HitFX that figure out xBABIP based on hardness of contact and batted ball type which can also adjust for quality of opposition. FIP can adjust for either bad luck or pitching worse than usual, but it cannot tell you why results are worse than they usually are.